Covariance and coherence matching for binaural room impulse responses; implementation and evaluation of rendering methods using different-order ambisonics

Abstract

Rendering binaural room impulse responses with the use of spherical microphone arrays and sets of head-related transfer functions minimize required measurements but will however exhibit a difference in perception due to, among other things, spherical harmonic truncation. By utilizing the spatial perceptually relevant interaural cues, the similarity to a directly measured BRIR could be enhanced. In this thesis, two different methods for rendering binaural room impulse responses with different-order Ambisonics are employed to evaluate and compare the similarity between the generated BRIRs and directly measured counterparts using an artificial head and torso recording. Both methods utilize inter-aural cues within the inter-aural coherence and covariance matrix, respectively, in order to process the diffuse part of the BRIR. The results show an improvement for the covariance matrix framework compared to the coherence matching method, even for rendering employing first-order Ambisonics. A pilot study in the form of an informal listening test was conducted to investigate the perceived similarity between the generated and measured BRIRs. The results of the listening test indicate that an increased spatial resolution due to higher-order Ambisonics will enhance the perceived similarity, while also displaying varying results depending on rendering method. However, a more extensive listening test would be beneficial. The findings indicate that the covariance framework with even higher-order Ambisonics would further optimize the quality and perceived similarity.